Artem Voloskovets
It connects three parts of the development process that are often evaluated separately:
- quantum algorithms;
- quantum error correction;
- physical hardware requirements.
The goal is to calculate how software and error-correction choices translate into the number and characteristics of physical qubits required to run a workload.
What Changed in Practice
The results cited for CUDA-Q Logical come from two separate projects: Fermilab measured development time, while Iceberg Quantum and Diraq modeled qubit requirements.
| Organization | What was measured | Previous / comparison point | Result with CUDA-Q Logical | Reported change |
| Fermilab | Architecture-development workflow time | ~5 months | ~3 weeks | ~7× faster |
| Iceberg Quantum + Diraq | Physical qubits estimated for a 1,000-logical-qubit model | Diraq’s previous estimate | ~150,000 physical qubits | ~10× fewer |
| Iceberg Quantum + Diraq | Logical qubits represented in the resource model | — | 1,000 logical qubits | Simulation target, not operating hardware |
The 7× figure belongs only to Fermilab. It describes the reduction in time required for one architecture-development workflow, from roughly five months to three weeks. It is not a 7× increase in QPU performance.
The 1,000 logical qubits, ~150,000 physical qubits and ~10× reduction come from work by Iceberg Quantum and Diraq. The 10× comparison is against Diraq’s previous physical-qubit estimate, not against Fermilab or another quantum computer.
- Fermilab: ~5 months → ~3 weeks.
- Iceberg Quantum + Diraq: 1,000 logical qubits → ~150,000 estimated physical qubits.
The latter is a simulated resource estimate, not a working system with 1,000 logical qubits. Actual hardware requirements depend on physical error rates, connectivity, error-correction codes and QPU architecture.
QUOPS Adds a Cross-Platform Metric
CUDA-Q is also getting QUOPS, a hardware-agnostic benchmark developed by Sandia National Laboratories.
QUOPS is separate from the Fermilab and Iceberg Quantum/Diraq results. It is intended to compare progress toward utility-scale quantum computing across different hardware platforms.
That distinction matters because the industry is developing competing approaches, including:
- superconducting qubits
- trapped ions
- neutral atoms
- silicon spin qubits
Physical qubit count alone does not show how much useful fault-tolerant computation a system can deliver.
Where NVIDIA Fits
CUDA-Q Logical is currently being used by Fermilab, Sandia, Infleqtion, IQM, QCDesign and Quantum Motion. This does not mean the performance figures above apply to all six organizations.
NVIDIA is building the layer connecting quantum processors with classical accelerated computing:
Quantum processor → CUDA-Q → NVIDIA GPUs → classical HPC
Fault-tolerant quantum systems require classical processing for error correction, control and hybrid workloads. CUDA-Q gives NVIDIA a software layer that can work across different QPU architectures rather than depending on one qubit technology.
So far, the concrete results are narrowly defined: Fermilab reports ~7× faster architecture-development workflow, while Iceberg Quantum and Diraq report a model requiring ~150,000 physical qubits for 1,000 logical qubits — ~10× below Diraq’s previous estimate.
Artem Voloskovets
